1 Adsorption of NO 2 on Pd(100) Juan M. Lorenzi, Sebastian Matera, and Karsten Reuter,

Transcription

1 Supporting information: Synergistic inhibition of oxide formation in oxidation catalysis: A first-principles kinetic Monte Carlo study of NO+CO oxidation at Pd(100) Juan M. Lorenzi, Sebastian Matera, and Karsten Reuter, Chair for Theoretical Chemistry and Catalysis Research Center, Technische Universität München, Lichtenbergstr. 4, Garching (Germany), and Fachbereich f. Mathematik u. Informatik, Freie Universität Berlin, Otto-von-Simson-Str. 19, D Berlin (Germany) karsten.reuter@ch.tum.de 1 Adsorption of NO 2 on Pd(100) Figure S1: Metastable adsorption configurations found for NO 2 on Pd(100). Energies are relative to the most stable configuration shown in the main text. Several adsorption geometries for the NO 2 molecule were analyzed within our densityfunctional theory (DFT) setup to determine the most stable configuration. For this, the molecule (in its vacuum-converged geometry) was placed on different high symmetry positions over a 4-layer thick, (2 2) Pd(100) slab. These configurations were used as initial conditions for geometry optimizations, in which the molecule and the upper two layers of the substrate were allowed to relax. The initial configurations included: (i) upright configurations with the N atom down over each of the high symmetry sites (top, bridge and hollow) and with the O atoms aligned either parallel, at a 45 angle or (only for the bridge site) perpendicular to the Pd-Pd direction; and (ii) tilted configurations, one for each of the upright ones, in which the molecule was rotated 45 around the N atom along an axis perpendicular to its plane, so that one of the O atoms was closer to the surface. After the geometry optimizations most of the initial configurations relaxed to the geometry presented in Fig. 1 of the main text, the one with the lowest energy observed. The four next most stable final geometries found are presented in figure S1 along with their relative energy with respect to the most stable configuration. To whom correspondence should be addressed Technische Universität München Freie Universität Berlin S1

2 Table S1: Lateral interaction parameters per adsorbate for the selected lattice-gas Hamiltonian. Values in mev. V 1NN O O CO CO CO CO O NO O CO 2 NO 2 NO O O O NO 2 V 3NN V 3NN Lateral Interactions To compute the strength of the lateral interactions and guide the definition of site-blocking rules of our 1p-kMC model, we calculated a database of DFT energies for different overlayer configurations. Adsorbates were placed on their corresponding adsorption site in supercells containing slabs with either (2 2) or (3 2) surface unit cells and 4 layer thickness. Preliminary calculations showed that configurations containing any pair of adsorbates at distances closer than the first nearest neighbor (1NN) distance between top and hollow sites ( 1.97 Å) do not correspond to potential energy minima (i.e. adsorbates moved away from the original adsorption sites during geometry optimizations). Consequently, no configuration with coadsorption at such close distances was considered. In all configurations, O, CO and NO adsorbates and the topmost two substrate layers were allowed to fully relax during the optimizations. Aiming to describe higher coverage conditions, NO 2 was constrained to maintain a C 4v symmetry above the top site. Under this constraint NO 2 -NO 2 interactions at 1NN top-top distance turned out extremely large and tended to corrupt the inversion of the lattice gas Hamiltonian (LGH). We correspondingly concluded on a site-blocking rule for this short distance and did not consider corresponding configurations in the database. Configurations for which any Figure S2: Representation of the interactions selected for the lattice gas Hamiltonian from table S1. XO and YO represent either CO or NO. of the adsorbates was displaced to a different adsorption site were also discarded and not considered in the analysis below. Figure S5 at the end of the document schematically shows the resulting 97 DFT calculated configurations that were used. Binding energies at the zero coverage limit were taken from calculations with a single adsorbate in (3 3) surface unit cells (0.11 ML coverage). The DFT energies obtained for these configurations were used to fit a short-ranged LGH. Due to the relatively large number of adsorbates, we considered pairwise interactions only. Starting with a set of 23 interaction figures, and following the approach used by Zhang et al., S1 we found that the most accurate LGH (i.e. that which minimizes the leave-one-out cross validation (LOO-CV) score S1 ) corresponds to the one presented in table S1, in which the 16 interactions shown in Fig S2 were selected. However, even when only allowing a maximum of 12 interactions, the strongest interactions in table S1 are selected with values that differ by less than 20 mev from the corresponding values in table S1. Additionally, we have also performed the same analysis by excluding all configurations (and interactions) containing NO 2. Under these conditions we observe that again the 9 strongest NO 2 -free interactions from table S1 are selected, and that the corresponding values differ again by less that 30 mev. We are thus S2

3 confident that the symmetrized treatment given to NO 2 does not have important effects on the interactions between the other species at the level of accuracy we aim at in this analysis. The highly repulsive interaction values obtained for the shortest-range distances in table S1 and the robustness with which they are obtained, fully justifies the site-blocking rules employed in the 1p-kMC model in the main text. 3 Influence of the blockingrules scheme The set of blocking rules used was defined such that all interactions stronger than 130 ev per adsorbate would be blocked. This corresponds to prohibiting (cf. table S1) a. O-O coadsorption at 1NN hollow-hollow distances b. CO-CO, CO-NO and NO-NO coadsorptions at 2NN bridge-bridge distances across top sites, and c. CO-CO coadsorptions at 2NN bridge-bridge distances across hollow sites in addition to blocking coadsorptions at distances equal or smaller than 1.97 Å (which do not even correspond to metastable configurations as discussed in the previous section). This was done to keep our extended set of blocking rules fully consistent with the one used by Hoffmann and Reuter. S2 The robustness of our results with respect to this choice has been systematically checked as follows: Fig. S3 compares the TOF for CO oxidation predicted with the blocking-rules scheme just described (solid lines, same data as in Fig. 3 of the main text), with the predictions from two alternative schemes, one softer (dashed lines) and one harder (dotted lines). The softer model corresponds to releasing the blocking of CO-CO coadsorption at 2NN bridge-bridge distance across hollow sites (effectively raising the blocking cut-off energy to 170 ev). It can be seen that the effects are completely negligible. On the other hand, the harder scheme extends blocking by prohibiting all CO-CO, CO- NO and NO-NO coadsorptions at 2NN bridgebridge distances (i.e. both across hollow and top sites). Although somewhat larger than in the previous case, differences are still small. The most notable deviation is observed for the poisoned case, and accounts to a shift of the transition in the p NO axis. The farther reaching interactions included into the site-blocking of this harder scheme have only smaller finite values. In reality, adsorption into corresponding configurations will not be entirely blocked. We therefore expect the effect predicted by the harder scheme to actually represent an upper bound, i.e. the true quantitative effect of neglecting the finite farther-reaching repulsive interactions in the main model will be less than what is predicted by the harder scheme. All in all, this analysis thus shows clearly that none of the qualitative effects put forward in this work are affected by the details of the employed blocking-rule scheme. TOF [site 1 s 1 ] Figure S3: TOF as a function of NO partial pressure and at T = 600 K for the high activity (magenta, left panel) and poisoned (black, right panel), predicted using different blocking-rule schemes. Solid lines correspond to the results reported in the main text (cf. Fig. 3); dashed lines, to a softer lateral interaction model; and dotted lines, to harder one (details in text). S3

4 4 Results with modified process lists TOF [site 1 s 1 ] Figure S4: CO oxidation TOFs at T = 600 K for the high activity (magenta, left panel) and poisoned (black, right panel) cases as in Fig. 3 of the main text. Compared are results from three altered 1p-kMC models: Dotted lines correspond to the model with NO oxidation completely turned off. Dashed lines correspond to a model in which NO 2 formation (but no desorption) is allowed, but dissociation is restricted (see text). Solid lines correspond to the model in which NO 2 formation and dissociation is unrestricted (but NO 2 desorption is turned off). As discussed in the main text, we have analyzed different modifications to our 1p-kMC model to clearly elucidate the source of the reactivity enhancement. The modifications implemented are: (i) the NO oxidation elementary process was completely turned off; (ii) the surface NO oxidation process is on, but the NO 2 desorption process is turned off and the NO 2 dissociation process is restricted, such that this intermediate can only dissociate into the NO + O configuration that initially generated it; and (iii) only NO 2 desorption is turned off and NO 2 formation and dissociation is unrestricted. The difference between cases (ii) and (iii) allow to isolate the effects of the NO 2 mediated O mobility schematically shown in Fig 5 of the main text. The resulting CO oxidation TOF for each of these cases is presented in Fig S4, with results from cases (i), (ii) and (iii) shown with dotted, dashed and solid lines, respectively. As discussed, for the poisoned case the incremental addition of the different elementary reaction events shifts the position of the transition from O-poisoned to the NO+O reactive state to lower values of p NO. 5 Kinetic Monte Carlo vs. mean-field microkinetics Mean-field microkinetics is generally not able to properly model a system with strong lateral interactions. Following the approach by Temel et al. S3 we compared the results of our 1p-kMC model to a mean-field description based on exactly the same elementary process list and exactly the same first-principles based rate constants. In complete analogy to the work by Temel et al. S3 for CO oxidation at RuO 2 (110) we found the mean-field description to predict a much wider pressure range of appreciable catalytic activity at T = 600 K already in the absence of NO. In fact, no O-poisoned regime is obtained in the entire pressure range shown in Fig. 2 of the main text. With this deficiency the mean-field model would correspondingly not be able to reliably capture the extended activity range observed in the 1p-kMC model at finite NO pressure. References (S1) Zhang, Y.; Blum, V.; Reuter, K. Phys. Rev. B 2007, 75, (S2) Hoffmann, M. J.; Reuter, K. Top. Catal. 2014, 57, (S3) Temel, B.; Meskine, H.; Reuter, K.; Scheffler, M.; Metiu, H. J. Chem. Phys. 2007, 126, S4

5 Figure S5: Configurations used for calculating the lateral interaction parameters. Gray circles represent Pd substrate atoms. Red, black, blue and green circles represent O, CO, NO, and NO 2 adsorbates, respectively. S5